Table 2.

Comparison of RegioSQM to two ML-based models. Tautomers are not considered so “RegioSQM20*” (note the “*”) corresponds to the “FF Optimization” entry in Table 1

Methodology	Full data set	One reactive atom
	Corr/Semi/Fail	Corr/Fail
RegioSQM20*	90.7%/5.4%/3.9%	92.0%/8.0%
WLN	89.1%/4.9%/6.0%	96.0%/4.0%
IBM RXN	76.3%/23.7%	85.0%/15.0%

“One reactive atom” refers to the subset of the full data set with only one unique experimentally observed reaction site and where RegioSQM20* only predicts one unique (green) reactive site